CPU3DAI tools for 3D, video and audio

Inkling-2-Mini text-generation model appears on HuggingFace

On November 17, 2026, the model thinkingmachines/Inkling-2-Mini was published on HuggingFace with the text-generation task. In the first hours, the model collected 819 likes with 136 downloads. The code has been released as open source.

What it means

The model contains 26.4 billion parameters and uses the MoE architecture. The largest weights file is 18.2 GB in safetensors format; the repository totals 18.8 GB. The license is Apache 2.0, and it runs via transformers. For local runs, the platform already hosts 35 quantized builds: GGUF, NVFP4, FP8, and MLX. The most compact working build — IQ4_XS by AtomicChat — takes up 14 GB. This makes it possible to run the model on GPUs with the corresponding amount of VRAM without additional modification. The model has only just appeared, so there are no practical reviews of generation quality yet. You can follow news about tools for local runs in the CPU3D news feed.
Inkling-2-Mini on HuggingFace November 17, 2026 · text-generation · Apache 2.0 Model thinkingmachines/Inkling-2-Mini 26.4B parameters MoE architecture text-generation task Weights safetensors format Largest file: 18.2 GB Repository total: 18.8 GB runs via transformers Quantization 35 builds GGUF · NVFP4 · FP8 · MLX IQ4_XS by AtomicChat — 14 GB for local runs Local runs on a GPU 14 GB VRAM no tweaks needed compact build Reviews on generation quality No hands-on reviews yet The model just came out follow in the CPU3D news feed Stats in the first hours 819 likes 136 downloads open source
How the method works. The diagram is drawn based on this news note.

See also